A Hybrid Delay-aware Approach Towards UAV Flight Data Anomaly Detection

With the rapid development of unmanned aerial vehicle (UAV) technologies, UAVs are now increasingly leveraged to perform military and civilian tasks today. Meanwhile, as a complex cyber-physical system, UAVs are also facing security and reliability concerns raised by internal systems errors and external cyber-attacks from multiple aspects. Recent research has spent efforts on leveraging AI and machine learning techniques to predict the flying status of UAVs using their flight data for anomaly detection. However, these methods often ignore the prediction delay existing in status-changing periods during the UAV’s operation, which inevitably causes false alarms and opens a window for malicious adversaries if they are not appropriately addressed. In this paper, we propose a new approach to enable effective anomaly detection and recovery for UAV flight data. Our approach adopts a hybrid design to eliminate false alarms during the status-changing periods while maintaining the high reliability of anomaly detection. We evaluate the proposed approach on flight data collected from multiple UAV flight paths. Our evaluation results validate the effectiveness of our hybrid design, which achieves both high anomaly detection accuracy and reliable recovery.